# Pseudo-colour ECG images boost AI heart‑risk detection and explanations

> Adding a clinically designed colour map to ECG images lifted AUC from 0.72 to over 0.9 and made model explanations more stable in a UK‑Biobank epilepsy cohort.

Oossa · 2026-11-10 · https://oossa.com/en/pseudo-colour-ecg-images-boost-ai-heart-risk-detection-and-explanations

Researchers trained two identical residual networks on single‑heartbeat ECG images. One network received a pseudo‑colour encoding that fills the area under the QT interval with colours chosen from the QT‑nomogram, a clinical risk tool. The colour‑only model reached AUC scores of 0.903 and 0.914 on two UK‑Biobank visits, while the monochrome model stayed around 0.72. Across eight explanation techniques, the coloured inputs gave clearer and more stable localisation of the risky QT region.

## The facts

- AUC 0.903 and 0.914 for coloured model vs 0.719 and 0.720 for monochrome
- Study published 10 Nov 2026

## Why it matters

Clinicians could rely on AI‑generated ECG explanations that align with known electrophysiology, improving trust in automated risk assessments.

## Sources & references

1. [Clinically Informed Colour Encoding Improves the Localisation and Stability of Explanations in Deep Learning ECG Models: External Validation in a UK Biobank Epilepsy Cohort](https://arxiv.org/abs/2611.09971) – arXiv cs.CV, 2026-11-10

Last updated: 2026-11-10